feat(gpu): optimize packing keyswitch on gpu

[andrei-stoian-zama]

GPU optimization for packing keyswitch for any level count using GEMM.

Bench using make bench_integer_compression_gpu on GTX 4060 mobile

Bench	timing ms	latency difference w.r.t previous impl
cuda::packing_compression::pack_u2	1.6173	-35.22%
cuda::packing_compression::pack_u8	1.6211	-31.76%
cuda::packing_compression::pack_u16	1.6316	-32.55%
cuda::packing_compression::pack_u32	1.6345	-38.88%
cuda::packing_compression::pack_u64	1.648	-70.32%
cuda::packing_compression::pack_u128	1.7274	-81.30%
cuda::packing_compression::pack_u256	3.3431	-81.69%
cuda::packing_compression::pack_u512	5.9654	-84.47%